| --- |
| language: |
| - ru |
| license: mit |
| tags: |
| - text-generation |
| - pytorch |
| - qwen2 |
| - russian |
| - tensor |
| pipeline_tag: text-generation |
| --- |
| |
| # Tensor-2-40m-base |
|
|
| Tensor-2-40m-base is a Russian-language language model from the **Tensor** series, developed as part of the **GribAI** project. This is a base (pretrain) model without instruction tuning. |
|
|
| ## Description |
|
|
| Compared to previous models in the series, Tensor-2-40m-base shows noticeably better text continuation and stronger understanding of the Russian language — both grammatically and in terms of semantic coherence between sentences. |
|
|
| ## Training |
|
|
| The model was trained on **70 MB** of Russian-language text data. |
|
|
| ## Usage |
|
|
| ```python |
| from transformers import AutoModelForCausalLM, AutoTokenizer |
| |
| model_name = "VGribAI/Tensor-2-40m-base" |
| |
| tokenizer = AutoTokenizer.from_pretrained(model_name) |
| model = AutoModelForCausalLM.from_pretrained(model_name) |
| |
| prompt = "Привет, как дела" |
| inputs = tokenizer(prompt, return_tensors="pt") |
| output = model.generate(**inputs, max_new_tokens=100) |
| |
| print(tokenizer.decode(output[0], skip_special_tokens=True)) |
| ``` |
|
|
| ## Limitations |
|
|
| This is a base (pretrain) model with no dialogue alignment — it's meant for text continuation, not for answering questions or following instructions. |
| **GribAI** project (VGribAI). |
